Development of a Smart Sensor for Acetone Detection Using Long-Period Fiber Grating and Machine Learning

Long-period fiber gratings (LPFGs) are widely used for the development of different types of sensors. It has recently been demonstrated that the use of Principal Component Analysis (PCA) can greatly simplify the analysis of the response of this type of sensor. By applying PCA to the transmission spectra of LPFG, the relationship between them and the magnitude of the physical variable of interest can be clearly observed. On the other hand, machine learning (ML) algorithms have proven to be very useful in identifying patterns and data regressions. This paper presents the characterization of an LPFG-based acetone sensor and its study using PCA as a dimension reduction technique, as well as the ML algorithm called support vector machine (SVM) as a multi-class classifier, to qualitatively and quantitatively determine the level of acetone concentration measured by the sensor. The results showed a reduction in the limit of detection (LOD) to 130 ppm at minute 15, as well as earlier detection at the first minute, with an increase in the LOD as a trade-off.